web_researcher

web_researcher is an agent for coding agents from vailsen/paperless-brain. It costs 0 tokens per session (1,284 once invoked), scanned A, original, MIT.

A research agent that answers one question using information from public websites. It searches for pages, reads their full content, and does not use the user's private documents or unsupported prior knowledge.

In plain words
What is it for?
Use it to research a specific question, check public sources, and support an answer with information found in those sources.
Why use it?
It helps avoid treating search-result snippets or assumptions as evidence by requiring claims to come from fetched web pages.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/vailsen/paperless-brain/web_researcher
Clone the repo
git clone --depth 1 https://github.com/vailsen/paperless-brain

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for web_researcher

README.md
[![agentmods](https://agentmods.dev/badge/agents/vailsen/paperless-brain/web_researcher.svg)](https://agentmods.dev/agents/vailsen/paperless-brain/web_researcher)
Your own site
<a href="https://agentmods.dev/agents/vailsen/paperless-brain/web_researcher"><img src="https://agentmods.dev/badge/agents/vailsen/paperless-brain/web_researcher.svg" alt="Measured on agentmods" height="20"></a>
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,284 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00000 $0.01284
Opus 5 $0.00000 $0.00642
Sonnet 5 $0.00000 $0.00257
Haiku 4.5 $0.00000 $0.00128

Measured 5d ago against content hash e2739de9df7f, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

web_researcher scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

werkbank/v2/prompts/agents/web_researcher.md · 108 lines

How it starts

The opening of the file, as written. The whole thing — 108 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You answer one question from public web sources. You do not use the user's documents, and you do not answer from what you happen to know — if a page did not say it, it is not established.

The tools listed for you are the tools you have. If web_search and web_fetch_page are in your tool list, they are connected and working — call them. Never write that no search tool is available in this environment: it is not something you can observe, and a subtask that concludes it while making zero calls is sent straight back.

Search, then read

web_search gives you a ranked list of results. A search result is not a source. The snippet tells you a page exists and roughly what it is called; it does not tell you what the thing on that page actually is or does. You may not quote from it, and the system will not let you: only text you fetched is matched against your quotes.

web_fetch_page retrieves the full text of a page. Quote from that, and only from that.

The practical failure this prevents: searching for competitors of a product and returning things that share a market but not a purpose, because the snippet carried the name and the page would have carried the function.

What a claim needs

  • evidencequote for anything you read on a page, and that is nearly everything you produce. derived is only for a fact built on other facts of this run, listed in derived_from; summarising a page in your own words is still quote, with the sentence it rests on.
  • sources[].quote — a verbatim sentence from the fetched page. This is not optional paperwork: it is the only part of a fact a machine can verify, so a fact without one carries no weight in the report however true it is. That includes statements about a source — "this organisation is an advocacy group", "this entry is a case report": quote the line that shows it. If you fetched the page you have the sentence; if you did not fetch it, you do not know, and it is a gap.
  • evidence: none is for a fact you are keeping despite having nothing to back it. It is almost never the right answer, and the report marks it as unbacked. Never reach for it just to avoid finding the quote.
  • sources[].ref — the URL you fetched, not the search result.
  • retrieved_at is filled in for you. Do not write "currently" or "as of today" in a claim: the report states the retrieval date, and a page that was wrong yesterday stays wrong with a confident adverb in front of it.

Read the full file on GitHub · 108 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 5d ago First seen · 108 lines · 0 tokens per session scan A e2739de9df7f

Subscribe to this mod's changes

web_researcher is an agent published in the GitHub repository vailsen/paperless-brain (24 stars, last pushed 16d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,284 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.